A robot may need information about a warehouse entrance, a road junction or a construction site it has never encountered. Someone nearby might be able to capture that environment with nothing more than a smartphone. That’s the idea behind @vangrid_io, which has raised $9M to develop a data network for Physical AI. Collecting this information at scale is difficult. Satellites cannot provide every ground-level detail, and sending specialist mapping teams to thousands of locations can be expensive and time-consuming. Vangrid offers another approach: let contributors capture requested places, then process those recordings into spatial data that AI developers can use. This is different from simply uploading videos. The goal is to produce useful representations of physical environments, including their layout and spatial relationships, for applications such as robotics, autonomous navigation and world models. There’s also an incentive for participation. Organisations can fund capture requests through bounties, while contributors can receive USDC for accepted work. Captures can be verified onchain, and payments are settled on Base. Vangrid reports more than 100,000 verified captures, alongside live bounties and USDC settlements. The long-term challenge is making this data broad, accurate and useful enough for real applications. A distributed network could help cover locations that traditional mapping methods miss, but the quality of the resulting data will matter just as much as the number of captures. Physical AI needs information about the world beyond controlled labs. @vangrid_io is working on a way to collect it from the ground up.